topic Re: Repairable system failure prediction in Discussions

Repairable system failure prediction

MarkovVaribles1 — Tue, 09 May 2023 19:59:40 GMT

Hello,

Here is a general question for JMP experts:
say I have data on a system that fails with various failure modes. The data contains dates of repair on each components repair. I I know how to analyze this data for reliability as shown in video here:
https://community.jmp.com/t5/Mastering-JMP/Analyzing-Reliability-for-Repairable-Systems/ta-p/483465

However what I am interested is in predicting the time at which the system will fail in the future and the probability of that failure.

Any idea/suggestion on how to do that in JMP?
Thanks,

Re: Repairable system failure prediction

Jed_Campbell — Wed, 10 May 2023 15:47:30 GMT

I'm not an expert in repairable systems, but I'll offer two suggestions as a catalyst for further discussion (and to keep your post from falling off the front page with 0 replies):

Use the MTBF and Failure Intensity Profilers in the Reliability Growth platform as proxies for "time at which the system will fail and probability of failure". This link suggests that this is standard practice for repairable systems.
(Perhaps not as statistically rigorous) If you hide and exclude any rows in your data that include 0 fixes, then you could treat the data as a non-repairable system and use the Life Distribution Profilers to model time (Distribution, Quantile, Density) and probability (Hazard Profile) of failures.

I've attached a sample dataset with scripts for both of these approaches saved to the table.

Re: Repairable system failure prediction

Mark_Bailey — Wed, 10 May 2023 18:18:06 GMT

To clarify, after modeling the lifetime data, the model will predict the time at which the probability of failure is a given level or the probability of failure at a given time. Both predictions are generally extrapolations beyond observed events. They incur wide confidence intervals.

Re: Repairable system failure prediction

OPersaud — Wed, 23 Jul 2025 12:22:32 GMT

Hello JMP Team,

Has there been a more direct method to compute this "time to next failure"?

Given, that we have outputs such as growth and scale parameters from NHPP or Crow AMSAA models?

Thanks.